The vocalic system of the Mehri of Oman

Stress, length and syllabic structure

In: Brill's Journal of Afroasiatic Languages and Linguistics
View More View Less
  • 1 UMR 6310 CNRS, Université de Nantes
  • 2 UMR 7110 CNRS, Université Paris 7

Existing grammars and descriptions of the Mehri of Oman usually deal with a presentation of the surface vocalic system only. Our aim in this article is to clarify the phonological structure of this system. We examine the nature of stress and its interaction with vocalic length and syllabic structure, and conclude that the vocalic system of the Mehri of Oman does not include any opposition of quantity at the phonological level. All superficially long vowels are the product of one of two processes: lengthening of open syllables under stress, and compensatory lengthening after the deletion of consonants in the coda. Lengthening under stress is constrained by a simple and widely attested condition: it can only take place in open syllables, and never in closed ones.


Existing grammars and descriptions of the Mehri of Oman usually deal with a presentation of the surface vocalic system only. Our aim in this article is to clarify the phonological structure of this system. We examine the nature of stress and its interaction with vocalic length and syllabic structure, and conclude that the vocalic system of the Mehri of Oman does not include any opposition of quantity at the phonological level. All superficially long vowels are the product of one of two processes: lengthening of open syllables under stress, and compensatory lengthening after the deletion of consonants in the coda. Lengthening under stress is constrained by a simple and widely attested condition: it can only take place in open syllables, and never in closed ones.

Introduction: interaction between stress, vocalic length and syllabic structure

Existing grammars and descriptions of the Mehri of Oman1 usually deal with a presentation of the surface vocalic system only.2 As a reminder, see (1) below:

1 Mehri of Oman: The surface vocalic system

article image

Beyond this point, a phonological analysis of this system is not available. The obvious difficulties3 of offering a view of the phonological nature of this system are due, we believe, to the fact that the nature of stress and its interaction with vocalic length and syllabic structure have not been clearly analysed. Our goal in this paper is to clarify these points.4

Johnstone (1975: 10) writes: “Long vowels occur only in stressed open syllables or stressed final-VVC syllables”. In this fundamental observation, vocalic length is closely related to the presence of stress on the one hand and to the syllabic structure on the other hand. It predicts that long and short vowels are distributed as follows in (2):5

article image

In Mehri, a final CVC syllable behaves like an open syllable (Johnstone 1975: 10, Rubin 2010: 29). A final CVCC syllable is a closed syllable. Within this framework, (2) is to be read as follows: 1. Long vowels only appear under stress in open syllables or final CVC syllables (greyed cell); 2. Short vowels appear in unstressed syllables in all syllabic contexts and, under stress, only in closed syllables.

We believe that Johnstone’s statement is fundamentally correct and that the distribution of vowels in Mehri corresponds to (2).

Alternations under stress clearly point to (2). In Mehri, the length of the stressed vowels depends on the syllabic context: ū and ī in open syllables alternate with ə in closed syllables, and ē or ō with a under the same syllabic condition (ML: xiv):

article image

Beyond this point, to validate (2), it is necessary to establish that the configurations below are excluded:

4 stressed vowels:

  1. i.(short) v in open syllables (or in final CVC syllables)
  2. ii.(long) in closed syllables

unstressed vowels: (long) v̄

  1. open syllables
  2. closed syllables

On initial examination, the facts do not confirm these predictions: every single excluded configuration in (4) is attested in ML. Thus, one may notice the presence of:

article image

The offending forms can be grouped in two sets:

5 forms with unexpected long vowels:

  1. unstressed long vowels in open syllables
  2. long vowels, stressed and unstressed, in closed syllables

forms with unexpected short vowels: short stressed vowels in open syllables.

We successively examine these two sets of forms in sections 1 to 3 and show that they do not contravene (2): they are only the mechanical consequence of two phonological processes: 1. a compensatory lengthening process concerning ʕ or the glides y and w and 2. a latency process concerning ʕ. Then (in section 4), we analyse distribution (2) in the CV-framework and show that the Mehri vocalic system instantiates a configuration that is attested in a wide variety of typologically and genetically different languages. In section 5, we finally consider some particular issues that the vocalism of word-final syllables raises, in particular the lack of long unstressed vowels in absolute final position. We conclude that the vocalic system of the Mehri of Oman does not include any opposition of quantity at the phonological level.

1. Unstressed long vowels in open syllable

Johnstone (1975: 10), after having written: “Long vowels occur only in stressed open syllables or stressed final-VVC syllables”, adds: “it is possible however for a form to have more than one stressed long vowel”. Likewise, in the introduction of ML (xiii), after having pointed out that long vowels only appear in open stressed syllables (“In M a long vowel can occur only in an open stressed syllable”), he notes: “A form may however have two or more such vowels in a series of stressed open syllables, as, e.g. ʔāfīrūr, to become red”.6

In fact, there are many words with several long vowels in the Mehri of Oman. This, however, does not imply that they are all stressed. As far as facts are concerned, Rubin (2010: 30), who has consulted the recordings of Johnstone, clearly doubts the stressed character of those multiple long vowels: “it does not seem to be the case that all long vowels in a word with multiple long vowels are stressed”. Our recordings of native speakers follow the same trend. And, really, there is no sense in thinking that several stressed vowels could coexist in a word. By definition, stress prominence can only be unique: a word has one, and only one, stressed vowel.7

Therefore, we exclude this « explanation » of multiple long vowels and show that in fact:

6 long unstressed vowels in open syllables arise as a consequence of compensatory lengthening phenomena triggered by the loss of specific consonants in coda position: ʕ, w, y and marginally ʔ.8

1.1. The case of ʕ

Johnstone notes (ML: xiii): “ā often indicat[es] the presence of a non-explicit ʕ”. As shown in the examples in (7), this remark is correct, but it must be made more precise. Indeed, ʕ surfaces in [ā] only when located in coda position. Following rule obtains: əʕC → āC.9

article image

In post-consonantal onset position, ʕ gets lost as well, but in this case there is no compensatory lengthening. Schwa is lowered to a, and a is not lengthened: CʕəC → CaC, e.g. /yəhagʕər/ → yəhagar (not *yəhagār) √gʕr h-F sbj 3ms throw. Likewise, if in intervocalic onset position of an unstressed syllable, ʕ gets lost and colors ə to a, but no compensatory lengthening obtains: /yəharbəʕən/ > yəharban √rbʕ h-F cd 3ms lift, pull up. We face here a potentially universal regularity (Hayes 1989: 281 and quoted references): compensatory lengthening only concerns coda consonants.

In word-initial position, the realisation of ʕ calls for a particular comment. In this context, ʕ is in onset—not in coda—position. Its loss thus should result in a short vowel, not in a long vowel. Now we observe a long ā (preceded by ʔ):

article image

It is likely that we have an anteposition of ə in these forms as is the case when R1 = l, r—more rarely m, n:10 #ʕəCC > #əʕCC. ʕ is then in coda position and long ā appears, as expected. These forms must be compared to t2-F with R1 = ʕ (9): in this case, ʕ is in coda position (/əʕtəCūC/) and the resulting form is ʔātəCūC, i.e. a form that is parallel to Fa ʔāmūr.

article image

To conclude, there are no underlying unstressed ā: [āC] < /əʕC/.11

1.2. The case of the glides

The phenomena involving the glides y, w are parallel: if y or w are in coda position, compensatory lengthening takes place, if they are in an onset position following a coda, the vocalization of the glide—when it occurs—results in a short vowel.12 This is true not only for radical glides but also for infixal glides. We examine those two cases in turn.

For radical y in coda position, the outcome is always ī: əyC → īC.

article image

For w, there is a higher degree of complexity. We still have a long vowel, but it can have three different colors: ū, ā (/ɛ̄) and ō, respectively illustrated in (11 a, b and c).13

11 əwC → ūC / āC / ōC

article image

For /-əwC-/ sequences, Johnstone generally does not point out any vocalization in ML. However, in ML: xliii n. 1, he indicates that a short u surfaces in həwrūd h-F pf 3ms take (beasts) down to water: “In general əw is realised as u”. Here we have an obvious problem of transcription. In the configurations exemplified (11a) that we elicited in our fieldwork, native speakers regularly produce -ū-, e.g. šūkūf: the vowel is long.

In the subjunctive of Fa derived from w-initial roots, ML gives forms with ā instead of ū: yāzēm for sbj 3ms give √wzm < /yəwzēm/, (11b). These forms are attested (cf. MT, e. g. lāzēm 48: l. 28, and our own data from fieldwork), but this realisation of /əw/ as ā is limited to a very precise morphological form: Fa subjunctive of w-initial roots.14 One may see it as an ancient lexicalized form that does not reflect out a phonological process at work in the language. Moreover, some of our (young) native speakers regularly produce a ū in these forms (this is a nice example of “regularization” by younger speakers, cf. Rubin (forth.): 18), as well as in all the other -əwC- sequences: yūzēm instead of yāzēm. It remains that, whatsoever its color, ā or ū, the long unstressed vowel results from a compensatory lengthening process triggered by the loss of an underlying w in coda position.

Finally, in the forms in (11c), we face a realisation of /əw/ as [ō]. This fact raises a complex question that goes beyond the subject of this paper, and we simply notice that these forms all derive from roots with R2 = w. The relevant point here is that the long unstressed vowel results from a compensatory lengthening process triggered by the loss of w in coda position.

Let us now consider the unstressed ī and ū that do not proceed from the vocalization of radical glides, but from infixal glides. These infixes are still poorly identified in the literature. We start with the verbal forms in (12).

12 infixed verbal forms

article image

Two arguments show that ī and ū in these forms are the surface realizations of underlying y and w.

First, in Omani Mehri, the ejectives θ’ s’ t’ ɬ’ š’ k’ and the gutturals χ ħ ʁ ʕ (infra C [+low]) regularly trigger the diphthongization of ī to ay and of ū to aw (as well as a lowering of ē to ā), cf. ML: xiv, Rubin 2010: 23–26, Bendjaballah & Ségéral 2014a etc. Now, the forms in (12) do not present any diphthongization in this context, e.g. ənħīrūr, not *ənħayrūr.15 One could imagine that this is due to the fact that the vowel is unstressed. But there is no reason why the effect triggered by C [+low] should not apply in unstressed position. We are then led to think that ī in ənħīrūr is not a vowel but the realisation of an underlying y consonant.

Second, the subjunctive forms of these verbs present an ay sequence if preceded by C [+low] (13a), but they also do so if preceded by C [–low]. The ay sequence cannot therefore be related to the presence of C [+low]: ay is not a diphthong, it is an a + y sequence. y appears as an unstressed ī in the perfective and in the imperfective because, in these forms, it is underlyingly preceded by ə: pf ənħīrūr is underlyingly /ənħəyrūr/.

The pf ənħīrūr is thus parallel to the pf of quadriliterals with n-, e.g. 4n-F √k’rbt’ ənk’ərbūt’: both are of the form əCCəCCūC. The subjunctive confirms the analysis: yəħwayrər is parallel to yənk’arbət’ (13c).16

article image

The nominal and adjectival forms with an unstressed ī17 are very similar. In these forms, unstressed ī does not diphthongize after C [+low]: it is just the realisation of a glide /y/ in coda position.18 We give a few examples in (14), all of the form C1əyC2ō/ēC3(-tən) (greyed cells).19

14 nouns and adjectives infixed by y: CəyCō/ēC(-tən)

article image

1.3. A marginal case: ā in nouns of ʔ-initial roots and the definite article ħ-

Sima (2002: 651–652 and nn. 14 and 15), relying on Johnstone (1970: pass., ML, MT), asserts that the prefixed definite article is ħ- for nouns deriving from originally ʔ-initial roots. For example:

article image

In the indefinite form, R1=ʔ is lost without leaving any trace—the reasons of this loss being unclear (for a possible interpretation, see Sima 2002: 660–661). In the definite form, on the contrary, an unstressed ā systematically appears after the definite article ħ. The same article ħ- is found as well in front of a small number of nouns with R1 = b, d, m, n, r, w, y (Sima 2002: 651), for example ħəmōh the water (MT 36 l. 27), but in these contexts ā never appears. In the forms in (15), the unstressed ā can only come, once more, from a compensatory lengthening process triggered by the loss of ʔ in coda position: /ħə+ʔbū/ > ħābūn. Therefore it is not the realisation of an underlying long vowel.

1.4. A problematic case: the diminutives20

Finally, a particular class of words is problematic. They are the nominal diminutives. As a matter of fact, all of them present an unstressed ā / ɛ̄ (Johnstone 1973).

article image

Nothing prevents us from thinking that these forms have an infixal consonant whose loss triggers compensatory lengthening like all we have seen previously. However we have neither an argument to validate this statement nor an hypothesis about the identity of this infixal consonant.

To sum up: long unstressed vowels never (with the possible exception of diminutive patterns?) are the realisation of phonologically long vowels. They result from compensatory lengthening phenomena triggered by the loss of a coda consonant. The consonants are: ʕ, ʔ (very marginally), w and y—the identity of the consonant in the case of the diminutives remaining to be made clear.

2. Long vowels, stressed and unstressed, in closed syllables

ML gives forms with stressed and unstressed long vowels in closed (non final) syllables.

These forms however are very few and limited to specific contexts: they are present always and only in forms starting with ʔā and derived from ʕ-initial roots.

As for verbs, in unstressed position (17a), all relevant forms are of the marginal t2-F type fətkūr: all in all, 5 forms, in which /ʕətCūC/ surfaces as ʔātCūC. In stressed position (17b), we have t1-F forms in which /ʕatCəC/ surfaces as ʔātCəC.21

17 long vowels in closed syllables, verbal forms22

article image

As far as nouns are concerned now, we give a representative sample of forms with a long vowel in closed syllables in (18a and b) (unstressed and stressed resp.)23

For unstressed vowels, the templates are CəCCēt, CəCCīt, CəCCēC and CəCCīC. In these forms, for ʕ-initial roots, the underlying sequence /ʕəCC-/ surfaces as [ʔāCC-]. For stressed vowels, the underlying template is /ʕaCC/, with a short vowel, and it surfaces as [ʔāCC].

18 long vowels in closed syllables, nominal forms

article image

As for adjectives finally, there are a few cases with an unstressed or a stressed long vowel (greyed cells in (19a) and (19b) respectively). In all cases, these forms derive from ʕ-initial roots.

19 long vowels in closed syllables, adjectival forms

article image

In all preceding examples, the process is the same as the one studied in 2.1.1: #ʕəC > #əʕC > #ʔāC. But in the case of the forms examined in this section, lengthening applies in a closed syllable, which does not conform to the expectation: #ʔəʕCC > #ʔāCC.

The transcriptions however are rather unclear.

First of all, for forms with long ā in a closed syllable, ML often gives forms with ə in the next syllable, that is to say with ā in open syllable: ʔāCəC-. This is true of all the verbal and nominal categories mentioned above. For t2-F, for example, there are 21 relevant forms with R1 = ʕ and 16 belong to the type in (20b).24

article image

Then, for t1-F, the transcriptions are even less stable: along the forms with long ā only, ML gives forms with two variants, ā and a (21a) and still others with a short a only (21b).

article image

Such a recurrent variation forbids giving any real credit to Johnstone’s transcriptions. Either the vowel is short, or, more probably, it is long, but it is in fact located in an open syllable (that is to say in the configuration analyzed above in 2.1.1) and not in a closed syllable. ā sometimes appears to be located in a closed syllable because the schwas are not consistently transcribed.

The reality of long vowels in closed syllables, stressed as well as unstressed, therefore remains to be proven. Even if they were really attested, they could not constitute an objection to the generalization in (2): if long vowels in closed syllables really existed, why would they appear in contexts involving an underlying initial ʕ, only? Obviously, the phonology of ʕ would be at stake here, and not the inventory of the syllabic structures of the language. There are no long vowels in closed syllables in Omani Mehri.

3. Short stressed vowels in open syllables

Mehri exhibits a certain number of short stressed vowels in open syllables (CVCV, CVC#),25 i.e., in a context where one expects long vowels.

3.1. The type θəbərkəm: phonotactics of ə

Forms like θəbərkəm or abaðərhəm (ML: xvii) present a short stressed vowel (ə or a) in an apparently open syllable. Johnstone clearly deals with this apparent anomaly: “Since -Cə—often gives -əC-, forms like abaðərhəm are very common. This does not constitute a re-opening of the syllable in terms of M phonology” (ML xvi n. 1). In other terms, a form like θəbərkəm results from a late phonetic process related to the phonotactics of schwa: the underlying form is /θəbrəkəm/ with regular ə in a closed syllable, like in other forms of the paradigm (pf 1/2ms θəbrək, 1/2d θəbrəki, 1p θəbrən, 3mp θəbrəm). Note that the pf 3ms with the suffixed dual object clitic presents the same phenomenon: rəfsəhi = /rəfəs-hi/ < rəfūs-hi √rfs he kicked them2 (ML: xvii).

3.2. Consonant latency

A similar optical illusion concerning the syllabic structure can be observed in two other cases: the nakak-type and the dəl-type. In these two cases, a consonant that is phonologically present does not have any phonetic realisation (it is “latent”). This absence on the surface gives the impression that the vowel is in open syllable.

All the forms like nakak derive from roots with R2 or R3 = ʕ. This consonant never surfaces as such. It is nonetheless present at the phonological level, as noted by Lonnet & Simeone-Senelle (1997): 354, “Le ʕ compte comme une consonne, même lorsqu’ il a disparu laissant pour trace une influence sur la voyelle (ouverture, allongement): yənakam (<yənakʕəm)”. Phonologically, the a of nakak is not in an open syllable but in a closed syllable.26

article image

Besides, the verbal, nominal and adjectival forms derived from biliteral roots are given in ML with a short vowel in CVC#27 (type dəl):

article image

However Johnstone is inconsistent in his transcription: in certain nouns, he notes the final geminate (√br abarr outside, √ɬr ɬar, ɬarr ill-health, evil etc.), the adjective s’əħ is realized as s’aħħ (Rubin forth.: e.g. 434, Text 36:2), and, as far as verbs are concerned, he adopts, at least in the introduction to ML, a transcription with a geminate whose second member is between parenthesis: dəl(l). We take this convention to be significant. If the form is suffixed by a vowel-initial morpheme, the second root consonant is systematically geminated in the surface: Fa pf 3ms dəl know but 3fs dəllūt, N s ɬad burden but ɬaddəhəm their (m) burden, Adj ms s’əħ alive, awake, healthy but fs s’əħħayt, etc.

For the same forms, the phonetic analysis of the data recently recorded on the field (2012–2015) with “young” native speakers (20–30 years old), always shows the presence of a final geminate [dəll]. The acoustical analysis shows that the final consonant of the verbs like dəll is systematically longer than its simplex counterpart. (In the case of stops, the spectrogram clearly shows that the final geminates are, as with simplex Cs, systematically released. The length of the occlusion then allows a comparison between geminate and simplex C in word-final position. Now, the analysis highlights a clear contrast of the duration of the occlusion between these consonants, cf. Bendjaballah & Ridouane 2016).28

The forms that Johnstone gives with a short stressed vowel in final CVC# syllables are thus in fact in closed /CVCC/# syllables. They do not contravene (2).

3.3. the forms in -Vyən.

A last group of forms presents the sequence -Vyən in final position, whereas we expect either -VVyən, or -Vyyən. These forms are of three types:

24 reflexive and reciprocal + suff. 1p:

ћənfayən, t’āt’īdayən resp. (Rubin 2010: 49, 51)

“forms in -ən” duals (= cd of all Forms and ipf of a-F, t2-F and š2-F):

type yərkəzayən

pf 1p from √CCv:

type hǝrsayǝn (h-F √rsv)

As for reflexive and reciprocal + suff. 1p (24a), after listening to the recordings of MT, we assert that these forms have -ɛyyǝn:

article image

The duals and pf 1p of √CCv with a short vowel are rare in MT: only one dual has been pointed out (26a) and the recording is unfortunately unavailable (Rubin, p. c.), and two forms of √CCv pf 1p with a short vowel (26b). After listening to the audio, we conclude that these forms do not to have a short but indeed a long vowel, patterning on a par with similar forms given in ML and other present in MT (26c).29

article image

On this point our own fieldwork recordings are unambiguous: we have -ɛyyǝn for the duals and -ɛ̄yǝn for the √CCv pf 1p:30

article image

Then there is no short stressed vowel followed by a simple y: *-vyǝn.

In short: at the phonological level, there are no short stressed vowels in open syllable in Mehri. Their presence on the surface, when they are not mistaken transcriptions (*-vyən for -yən / -vyyən), is only the effect of the phonotactics of əəbərkəm) or of the regular latency of certain consonants (nakak, dəl).

4. Recapitulation and analysis

Thus, at the end of the review and analysis of the facts, distribution (2), repeated below in (28), is validated:

article image

The fact that the few remaining problems (e.g. diminutives) are extremely limited indicates that they are not enough to invalidate this distribution at the phonological level.

As a consequence, a simple generalization on the nature of stress emerges: stress is a stress of duration. Mehri is a classic case of a language with vowel lengthening under stress, subject to a simple syllabic constraint:

29 a stressed vowel is lengthened

in a closed syllable, this lengthening is inhibited

Such a system can be captured without any particular difficulty within our theoretical framework, the CVCV model (Lowenstamm 1996, Scheer 2004) in Government Phonology (KLV 1990). In this model, the syllabic level consists of a strict alternation of onsets and nuclei (noted C and V). A consonant in coda position of a closed syllable is an onset followed by an empty nucleus (30a). A long vowel is a vocalic segment linked to two V-positions separated by an empty C-position (30b).

article image
article image

Stress is represented as a CV unit following the stressed vowel. This vowel spreads to the V-position of this CV-unit, and thus surfaces as a long V.

The representation of arōkəb (a-F pf 3ms √rkb “put on the fire”) for example is given in (31a) below. If the stressed vowel is in a closed syllable, for example in yarakbən (a-F ipf 3ms √rkb), the propagation of the vowel to the V-position of the stress CV-unit (in grey) is blocked (31b).

article image
article image

This blocking effect simply results from the application of a general principle, that of the licensing of V-positions:

32 a vowel can spread to an empty V-position only if this position is licensed

only a V-position that is identified by a segment can licence the empty V-position to its left; an empty V position cannot do it

In (32a), ə licenses the V-position of the inserted accentual CV: stressed o31 can spread to this position and it surfaces as a long vowel. On the contrary, in (32b), the V position to the right of the accentual CV, being empty, cannot license the V-position of the accentual CV: stressed a cannot spread.

5. The word-final syllable

A certain number of questions remain, which concern the word-final syllable.

5.1. CC#

The first question concerns the forms with a final CC# syllable, e.g. kətūb Fa pf 3ms √ktb write. In (2) and in the paragraph that follows, we had, following Johnstone (1975: 10), assimilated final closed syllables to open syllables. However ū in kətūb is in a position that corresponds a priori to (32b): the final nucleus, being empty, should not license the preceding V position (that of the accentual CV) and lengthening should be inhibited.

This comes within the framework of a well-known typology. In some languages, for example Egyptian Arabic, long stressed vowels are excluded in internal closed syllables but possible in word-final closed syllable. In other languages, for example Turkish, long stressed vowels are excluded in all closed syllables, final as well as internal. This typology is often attributed to a parametrical adjustment of the properties of the word-final empty nucleus: in some languages, like Egyptian Arabic, a word-final empty nucleus can license the preceding empty nucleus, in others, like Turkish, it cannot. Mehri would belong to the first of these two classes of languages. This parametrisation however is only an encoding of the observations.

To go beyond this stage, we propose that, in languages where the word-final empty nucleus licenses the one to its left, it is not empty at the phonological level: a vowel does exist and it licenses the V-position to its left. This vowel simply has no phonetic realisation—or, if one prefers, it has a ø allophone in the context /__#.

Now, there is sufficient evidence that this is precisely the case in Mehri. In Semitic, the final nucleus of pf 3ms was not initially empty. In Classical Arabic kataba and Geʿez qätälä, this final vowel is still present. We propose that the fact that the final empty V-position is a licensor in Mehri results from the presence of this vowel at the phonological level. Arabic or Geʿez and Mehri differ in that the realisation of the final V-position is ø in Mehri.32

If this analysis is correct, table (2) can and must be simplified. The word-final CVC# syllable being reanalyzed as a sequence of two open syllables CV.C(v)#, the distinction “CV / CVC#” becomes useless: vowel lengthening takes place in all stressed open syllabled.

article image

5.2. V#

If an underlying short unstressed vowel does not surface in absolute final position, as we claim, no short unstressed final vowel should be observed. Now, we frequently come across i#, a#, u#, ɛ#:33 rəkəzki pf 1/2d, θǝbrōna part ms, bīru Fb pf 3fp √brw bear, give birth, ħəbənhɛ his (m) sons respectively, for example.

The great majority of these short unstressed vowels are in fact the surface form of vocalized consonants: a# comes from /əʕ#/ (34a), cf. ML xiv n. 2: “Final a […] is confined to forms from root final ʕ, such as nūka”, i# from /əy#/ (34b)34 and u# from /əw#/ (34c):35

article image

A few V# examples, however, cannot be construed as vocalized consonants: they are vowels at the underlying level. The vowels and forms concerned are:

  1. a# in ms, fs and mp forms of the participle (35a);
  2. ɛ# in the Poss suffixed to a plural noun (35b);
  3. i# in 2fs ipf / sbj (35c.i) and in various markers of dual (35c.ii).
article image

For a# and ɛ# we do not see which consonant could have been lost.38 For i# of 2fs ipf / sbj (35c.i), one could imagine a consonantal origin but the vocalic (long) nature of the morpheme seems to be clear in Semitic, cf. Akk. taprusī, Hebr. tiqbərī, Ar. taqburīna. (For i# in duals (35c.ii), see below).

The crucial point here is that if the underlying short final vowels have no surface realisation, as stated in 5.1, these surface short vowels can only proceed from phonologically long vowels.

This phonological length finds a confirmation in the double treatment that the velar k underwent before a front high vowel. We have as a matter of fact a striking opposition between pf 2fs rəkəzš and pf 1/2d rəkəzki: in the first case we face a palatalisation of 2s -k before the feminine *-i, which then disappears: rəkəz+k+i > rəkəzš, and in the second case, we do not have any palatalisation before i, and this one remains stable. In fact, the difference of palatalising power, as well as the opposition between maintenance and disappearance, proceed from a difference in underlying length. The feminine *-i in the pf 2fs suffix is short (cf. Ar. katabti, Akk. parsāti): as such, in Mehri, it does not surface, but it “sets down” the velar, that is to say palatalises it. The -i of the dual, long and thus tied to two V positions (36a), cannot propagate to the velar, that is does not palatalise it, but on the other hand remains on the surface in the form of a short i (36b).

article image
article image

Thus, in Mehri, some final unstressed short vowels can be analyzed only as the surface form of underlying long vowels. In Mehri, /v/ > ø and /v̄/ > [v] / __#.

But these vowels are present only in four morphemes of the language (36a, b, c.i and c.ii) and we have shown in sections 2 to 4 that there is no quantitative opposition at the phonological level in the vocalic system of the Mehri of Oman. In fact, if it is clear that the final vowels in question were originally long vowels—and that their survival proceeds from there, we are bound to think that they are purely and simply lexicalized as short vowels in today’s language. The vocalic phonetic form that the final glides and ʕ regularly take (ɬīni, bīru, nūka) has possibly contributed to “install” these few short unstressed final vowels.

6. Conclusion

To conclude, it appears that there are no phonologically long vowels in the Mehri of Oman. All superficially long vowels are the product of one of two processes: lengthening of open syllables under stress, and compensatory lengthening after the deletion of consonants in coda position (ʕ, y, w, and, marginally, ʔ).

Lengthening under stress is constrained by a simple and widely attested condition: it can only take place in open syllables, and never in closed ones.

Diachronically, this means that the quantitative opposition of Proto-Semitic (Moscati & al. 1964: 46) has been lost. Other languages of the Semitic family exemplify a similar loss in various ways: Biblical Hebrew (Cantineau 1950: 118–120), Modern Arabic dialects, etc.

Based on the alternations in stressed syllables (and taking [ō] as the regular realisation of stressed /a/ in open syllables, cf. footnote 31), we can schematize the vocalic system of Mehri as follows:

article image

This leaves the question of vowel qualities. Each of the five vowels has multiple allophones. For example, stressed /e/ surfaces as

  1. [ē] in open syllables: /yərkez/ > [yərkēz] Fa sbj √rkz,
  2. [a] in closed syllables: /rəkezk/ > [rəkazk] Fπ pf 2ms √rkz
  3. [ā] in open syllables after C[+low]: /yərs’en/ > [yərs’ān] Fa sbj √rs’n
  4. [ɛ̄] before ʕ#: /yənkeʕ/ > [yənkɛ̄] Fa sbj √nkʕ.

A single surface vocalization can stem from various different underlying vowels: for example,

  1. stressed [a] is the realisation of
    1. /a/ in a closed syllable: /yarakbən/ > [yarakbən] aF ipf √rkb
    2. /e/ in a closed syllable: /rəkezk/ > [rəkazk] Fπ pf 2ms √rkz
    3. /u/ in a closed syllable followed by ʕ: /nukʕək/ > [nakak] Fa pf 2ms √nkʕ.
  2. unstressed [a] comes from /ə/ / Cʕ__: /nukʕək/ > [nakak]: Fa pf 2ms √nkʕ.
  3. unstressed [ə] comes from /ə/, but stressed [ə] can only come from /i/ or /u/ in closed syllables.

All of the three possible dimensions of conditioning (stress, syllable structure, segmental features, and combinations thereof) are involved in these realisations. A clarification of these conditioning properties must be left for a further study.


Fa, Fb, Fπ (passive) = Simple verbal forms: rəkūz, rīkəz, rəkēz (√rkz);

a-F, h-F, š1-F, š2-F, t1-F, t2-F = Derived verbal forms (a-, h-, š-, -t-): (a)rōkəz, (hə)rkūz, šərkūz, šərēkəz, ratkəz, ərtəkūz (√rkz);

F w ®, F y ®, n-F y ® = Verb forms with infixal -y/w- and reduplication of R3: ɬχəwlūl (√ɬχl), ɬ’əʁayʁūr (ɬ’ʁr), with n-: ənħəs’ībūb (√ħs’b);

4a-F, 4n-F = Quadriliteral forms with a-, n-: (a)k’arbət’, ənk’ərbūt’ (√k’rbt’);

Obj = Object pronoun, suffixed to the verb;

N = noun; N-īt, N-ēt = noun with suffix; Adj = adjective; Poss = possessive suffix;

m, f, s, p, d = masculine, feminine, singular, plural, dual;

pf, ipf, sbj, cd, part = perfective, imperfective, subjonctive, conditional, participle.


Bendjaballah, Sabrina & Philippe Ségéral (2013), “Remarques sur la gémination dans le système verbal du mehri (sudarabique moderne)”, in Phonologie, morphologie, syntaxe. Mélanges offerts à Jean-Pierre Angoujard (Ali Tifrit ed.): 31–60, Rennes: Presses Universitaires de Rennes.

Bendjaballah, Sabrina & Philippe Ségéral (2014a), “The Phonology of “Idle Glottis” consonants in the Mehri of Oman (Modern South Arabian)”, Journal of Semitic Studies LIX/1: 161–204.

Bendjaballah, Sabrina & Philippe Ségéral (2014b), “Le pronom objet en mehri d’ Oman”, communication at Colloque du réseau français de phonologie 12, Lille, 30.06–02.07 2014.

Bendjaballah, Sabrina & Philippe Ségéral (2017), “On the verb forms derived from four h-initial roots in the Mehri language of Oman”, Journal of Semitic Studies LXII/1: 199–215.

Bendjaballah, Sabrina & Rachid Ridouane (2016), “On word-final geminates in the Mehri language of Oman”, communication at Old world Conference in Phonology 13, Budapest, 13–16.01.2016.

Cantineau, Jean (1950), “Essai d’ une phonologie de l’ hébreu biblique”, Bulletin de la Société Linguistique de Paris 132: 82–122.

Fathi, Radwa (2015)a, “The architecture of Mehri Plurals (Modern South Arabian)”, Communication at The Form of Structure. The Structure of Form. Three days of Linguistics for Jean Lowenstamm, Université Paris Diderot Paris 7, 15–17.01.2015.

Fathi, Radwa (2015)b, “The many Omani Mehri plurals”, Communication at Journée d’ études sur les langues sudarabiques modernes 3, Strasbourg, France, 9.10.2015.

Fathi, Radwa (to appear), “Gender exponence and apparent polarity in a class of Omani Mehri plurals”. Glossa.

Hulst, Harry van der & Sam Hellmuth (2010), “Word accent systems in the Middle East”, in A Survey of Word Accentual Patterns in the Languages of the World (Goedemans, Rob W.N., Harry G. van der Hulst & Ellen van Zanten éds): 615–646, Berlin: Mouton de Gruyter.

Hyman, Larry (2006), “Word-prosodic typology”, Phonology 23: 225–257.

Johnstone, Thomas Muir (1970), “A Definite Article in the Modern South Arabian Languages”, Bulletin of the School of Oriental and African Studies (BSOAS), 33,2: 295–307.

Johnstone, Thomas Muir (1973), “Diminutive Patterns in the Modern South Arabian Languages”, Journal of Semitic Studies (JSS) 18/1: 98–107.

Johnstone, Thomas Muir (1975), “The Modern South Arabian Languages”, Monographic Journals of the Near East: 1–29 [/ Afroasiatic Linguistics (AAL) 1,5: 93–121].

KLV = Kaye, Jonathan, Jean Lowenstamm & Jean-Roger Vergnaud (1990), “Constituent structure and Government in Phonology”, Phonology Yearbook 7: 193–231.

Lonnet, Antoine & Marie-Claude Simeone-Senelle (1997), “La phonologie des langues sudarabiques modernes”, in Phonologies of Asia and Africa (Including the Caucasus) (Kaye, Alan S. ed.) vol. 1: 337–372, Winona Lake IN: Eisenbrauns.

Lonnet, Antoine (1985), “The Modern South Arabian Languages in the P.D.R. of Yemen”, Proceedings of the Seminar for Arabian Studies (PSAS) 15: 49–55, Londres (Seminar for Arabian Studies c/o Institute of Archeology).

Lonnet, Antoine (1993), “Quelques résultats en linguistique sudarabique moderne”, Quaderni di Studi Arabi (QSA) 11: 37–82.

Lonnet, Antoine (2005), “Quelques réflexions sur le verbe sudarabique moderne”, in Studi Afroasiatici: XI Incontro Italiano di Linguistica Camitosemitica (Mengozzi, Alessandro éd.): 187–201, Milano: Franco Angeli.

Lonnet, Antoine (2006), “Les langues sudarabiques modernes”, in Les langues chamito-sémitiques (afro-asiatiques), Faits de Langues n°27 (Amina Mettouchi & Antoine Lonnet eds): II-27–43, Paris: Ophrys.

Lowenstamm, Jean (1996), “CV as the only syllable type”, in Current Trends in Phonology: Models and Methods (Durand Jacques & Bernard Laks éds): 419–442, Manchester: ESRI.

ML = Johnstone, Thomas Muir (1977), Mehri Lexicon and English-Mehri Word-List, with Index of the English Definitions in the Jibbali Lexicon, compiled by G. Rex Smith, London: SOAS.

Moscati, Sabatino & al. (1980), An Introduction to the Comparative Grammar of Semitic Languages. Phonology and Morphology, Wiesbaden: Harrassowitz.

MT = Stroomer, Harry (1999).

Rubin, Aaron D. (2010), The Mehri Language of Oman, Leiden / Boston: Brill (Studies in Semitic Languages and Linguistics 58).

Rubin, Aaron D. (Forthcoming), The Mehri Language of Oman: Grammar and Texts.

Scheer, Tobias (2004), A lateral theory of phonology, Vol. 1: What is CVCV, and why should it be?, Berlin: Mouton de Gruyter.

Sima, Alexander (2002), “Der bestimmte Artikel im Mehri”, in “Sprich doch mit deinem Knechten Aramäisch, wir verstehen es!”, 60 Beiträge zur Semitistik: Festschrift für Otto Jestrow zum 60. Geburtstag (Werner Arnold & Hardmut Bobzin éds): 647–668, Wiesbaden: Harrassowitz.

Sima, Alexander (2009), Mehri-Texte aus der jemenitischen Šarqīyah, annotated and edited by Janet C.E. Watson & Werner Arnold, and in collaboration with ʿAskari Hugayrān Saʿd (Semitica Viva 47), Wiesbaden: Harrassowitz.

Simeone-Senelle, Marie-Claude (1997), “The Modern South Arabian Languages”, in The Semitic Languages (Robert Hetzron éd.): 378–423, London: Routledge.

Simeone-Senelle, Marie-Claude (2011), “Modern South Arabian”, in The Semitic Languages. An International Handbook (Weninger, Stefan éd. In collab. with Khan, Geoffrey, Streck, Michael P. & Janet C.E. Watson): 1073–1113, Berlin: De Gruyter.

Stroomer, Harry (1999) Mehri Texts from Oman. Based on the Field Materials of T.M. Johnstone, Wiesbaden: Harrassowitz.

Watson, Janet C.E. (2012), The Structure of Mehri, Wiesbaden: Harrassowitz.


We are grateful to Aaron Rubin for his comments on an earlier version of this article.


Mehri belongs, with Ḥarsūsi, Hobyōt, Jibbāli, Baṭḥari and Soqoṭri, to the Modern South Arabian language group, one of the branches of the Semitic family. Mehri is spoken in the east of Yemen and, in Oman, in the Dhofar governorate (see Johnstone 1975: 94, Lonnet 1985, Johnstone 1987: xi, Lonnet 1993, Simeone-Senelle 1997, Lonnet 2006, Sima 2009, Rubin 2010: 1–8). This paper deals with the Mehri of Oman. It is based on the data collected by Johnstone available in the Mehri Lexicon (Johnstone 1987, infra ML) and the Mehri Texts (Stroomer 1999, infra MT) as well as on the data that we recorded with native speakers between 2012 and 2015 within the frame of the OmanSaM project (ANR-13-BSH2-0001, In the examples, the Mehri forms in italics come from our own fieldwork, those in Roman from ML or MT.


See Johnstone (1975: 102–103), ML: xiii–xiv, Lonnet & Simeone-Senelle (1997: 362–363), Simeone-Senelle (1997: 385; 2011: 1081), Stroomer (1999: xxvii), Lonnet (2006: 30 n. 10), Rubin (2010: 22–23), Watson (2012: 22–32).


Rubin, in his grammar of the Mehri of Oman (2010: 22), notes that “Mehri vowels are a source of considerable disagreement” and Watson, in Sima (2009: 10), writes that “the vowels of the Modern South Arabian languages have exercised researchers for many years”.


The topic of word stress in Mehri has been reviewed in various papers, in particular Johnstone (1975: 10–12), ML: xiii, Lonnet & Simeone-Senelle (1997: 354), Simeone-Senelle (1997: 386), Lonnet (2006: 31), Rubin (2010: 29–30), Watson (2012: 34–35), Hulst & Hellmuth (2010: 625). These descriptions are facing numerous « exceptions » (Watson 2012: 34, Lonnet & Simeone-Senelle 1997: 354). We claim that this is due to the fact that they focalize on the question of the stress assignment rule and omit to define its nature. To quote Lonnet & Simeone-Senelle (1997: 354): “L’ accent tombe sur la (dernière) syllabe longue {CvC(C), Cv̄(C)} ou, si toutes les syllabes sont brèves {Cv}, sur la première (CvC en fin de mot compte pour brève)”, or Lonnet (2006: 31): “L’ accent porte en mehri sur la dernière syllabe longue (i. e. autre que CV ou CVC#) et sur la première si aucune n’ est longue.”


In (2) and further in the paper, the stressed vowel is underlined. We use a macron above the vowel to indicate vowel length: ū = [u:]. The Appendix provides a list of all abbreviations used in this article.


Johnstone (ML: xiii) nonetheless adds: “In most forms, however, there is only one long vowel”.


Hyman (2006: 231) formulates this principle as follows: culminativity: every lexical word has at most one syllable marked for the highest degree of metrical prominence”. In case there would be a so-called « primary » and « secondary stress », their status should be made clear, a somewhat clear assignation rule should be provided, and, first of all, the systematic link with vowel length and only with it should be justified.


l in the coda of a stressed syllable is also lost (səlmək > sɛ̄mək, see e.g. ML: xiii, Rubin 2010: 17). In this case however the process of compensatory lengthening cannot be seen as such, since it co-occurs with the lengthening of the stressed vowel. The presence of underlying /l/ can only be read off from the quality of the resulting vowel (ɛ̄).


In ML, notations with ɛ̄ can also be found, for example yəhɛ̄təmē < /yəhəʕtəmē/ h-F sbj 3md √ʕtm spend the night or mɛ̄θīt < /məʕθīt/ N-t fs goat droppings √mʕθ. In this article, we will mention these transcriptions but we will leave them aside: we believe this difference to be relevant at the phonetic level, only.


For example √lbd Fa əwbūd shoot, strike, √ltʁ Fπ əwtēʁ kill, √rs’n Fa ɛrs’awn tie, bind, √rfd Fa ɛrfūd give help to the poor, √mry Fa əmrōh, √nθ’f Fa ənθ’awf break a bone for its marrow, etc.


As for the glottal stop appearing in absolute initial position in Fa ʔāmūr, we come across it in t2-F ʔātəlūm. Now the presence of a vowel in absolute initial position does not trigger any ʔ-epenthesis in Mehri. In fact, in Fa, or t2-F, ʔ only appears with roots that have R1 = ʕ. We thus conclude that ʔ is not an epenthetic segment. It could be part of the surface realisation of /ʕ/ at the beginning of the word (in internal position, as previously seen, no glottal stop appears: ʕ has no consonantal expression, but only a vocalic one). These are probably the initial ʔ mentioned by Johnstone in ML: xii: “[W]hen [ʕ] occurs as a consonant it can usually be replaced by a glottal stop”. MT has w-’āmōr “and he said”, not *wə-’āmōr (MT 10: 4). This is unexpected: since ʔ is in intervocalic position, we expect wāmōr. However this latter form is in fact the one heard on the audio (Rubin (forth)). We therefore may conjecture that the ʔ that Johnstone systematically transcribes in initial position in Fa (8) and t2-F (9) from ʕ-initial roots is optional / a free variant of ø.


Usually, w and y do not get lost in onset position (t1-F pf 3ms √ʁyθ’ ʁatyəθ’ get angry, t1-F √ɬwl pf 3ms ɬatwəl look alike, etc). When they are vocalised, the result is a short vowel. The typical examples are: həʁigūt (/həʁyəgūt/) √ʁyg h-F pf 3fs (goats and sheep) bear young (ML: xlv n. 2) and (ML: xliv n. 2, xlv n. 2, lii n. 1 and 2) and hənufūt (/hənwəfūt/) √nwf h-F pf 3f beckon so. See also ML: xvi.


The forms in (c) need to be systematically confirmed with native speakers. A first check reveals that the form nōbēt is not attested, the correct form is nəbbēt, see also Rubin (forth, new MT 36:8).


Rubin (2010: 151) notes that these forms “behave as if the verb were of the I-ʕ type” and suggests that, like in other Semitic languages (e.g. Akkadian, Hebrew, Geʾez, Arabic), the sbj of w-initial verbs has no /w/. Note that a similar phenomenon obtains in one y-initial root: √ys’ (Fb ipf / sbj yās’ōs’ < /yəys’ōs’/). This example is, to the best of our knowledge, unique.


We nonetheless note ML Fy® √nt’r (ML √t’rr) ənt’ayrūr flow (blood) and √ɬ’ʁr (ML √ɬ’ʁr) ɬ’əʁayrūr scream, screech, shriek that contravene the generalization since they have a diphthong, not ī. We have not been able to check these two forms with native speakers. In MT these forms are realized with ī, ǝy or ay (Aaron Rubin p.c.) These various realizations may be interpreted as phonetic variants of /ǝy/.


Johnstone (ML: lxvii) says the same thing without justifying his position: he lists this verb class as “quinqueliteral verb[s] (or quadriliteral with infixed y)”. Moreover the infixal status of the glide in these forms is confirmed by the existence, at least in some cases, of forms deriving from an identical root, but without infixation. For example, for √hɬ’r, there is an infixed form: Fy® həɬ’īrūr go pale, green, yellow and non infixed forms: Adj ms həɬ’awr, fs həɬ’ərīt, p hēɬ’ər green, yellow and of N həɬ’ərāt green(ness). The same holds of √lbn, Fy® əwbīnūn go white (infixed) and Adj ms əwbōn, fs əwbənīt, p lēbən white and N əwbənēt whiteness (not infixed).


For ū, there is no clear case. Only three potential examples of N with an infixed w are listed in ML: nōmīl √nml ant, gōdēl √gwdl glowing stick (transcribed as gǝwdīl in Rubin (forth): MT 36: 8), ħəwt’ayk’ √ħwt’k’ curds. Johnstone gives a quadriliteral root with R2=w for the two last examples. The status of w, infixal or radical, is thus not clear in these forms. We leave this point at that. On infixal glides in the nominal system, see Fathi (to appear).


There are four exceptions in ML: two nouns (√ʁfn ʁayfēn coloured sheet used as a sari and √ħnv ħaynēʔ henna) and two adjectives (√k’s’r k’ays’ōr low, short and √ʁzr ʁayzōr deep) present ay. For a similar anomaly in verbs, cf. note 12.


This template is sometimes singular (14a.i), sometimes plural (14a.ii) in nouns, while it is uniformely plural in adjectives (14b). In the participle the /y/ infix unambiguously characterizes the plural of masculine forms (ms ktǝbōna ~ mp ktyēba). This suggests that this infix is a plural marker. The ms forms like t’əfays in (14b) call for a deeper analysis: they also have an infixed y but on a different site (Fathi 2015a, 2015b, to appear).


Besides the diminutives, that form an homogenous class, there are a few problematic nouns that exhibit a long unstressed vowel: mōk’ās’ piece of meat, k’ānūn law, nēwīw scarecrow, cloth laid over a tree to frighten wolves away and to keep o’s own animals nearby, ɬ’āfōr S Dhofar, nɛ̄t’ōr Ar. guard, ɬārēr water and yoghurt mixture, hārīk ʔarāk (?) tree, √ʔmn ʔīmān Ar. belief, faith, √ʔmr ʔɛ̄mīr Ar. prince, √ʔrχ (see √wrχ) tārīχ Ar. date … Most of them are loanwords (in particular from Arabic). We leave these few isolated words aside.


Let us also mention a unique 4a-F form, ʔārməd √ʕrmd (camel) run fast, < /aʕarməd/. It is very likely that the long vowel is in fact in open syllable, i.e. that we have a schwa after r. Indeed, besides ipf 3ms yārmōd (ML s.u.), Johnstone gives the form yārəmūd (ML: lxviii).


The pf 3ms alone is given in (17). The geminate surfacing in √ʕdw ʔāddō, √ʕðb ʔāððōb, √ʕsr ʔāssōr results from the assimilation of the t infix by the following R2 when it is “coronal”: /tR2/→ [R2R2] where R2 = [θ s t ɬ š θ’ s’ t’ ɬ’ š’ ð z d] (ML: xlvii, Lonnet 1993: 51, Lonnet & Simeone-Senelle 1997: 357, Rubin 2010: 110, Bendjaballah & Ségéral 2013: 35).


We may add a unique example of the form ʕaCCət: ʔāfyət √ʕfw good health, health, well-being, peace. Besides, we put aside a few isolated nouns derived from roots where the first consonant is not ʕ and for which ML gives long vowels in closed syllable: fāydət, faydət f √fyd advantage, benefit, k’ēʕyōt f √k’wʕ female spirit, ħāgērtən f √ʔgr (female) slaves, dāhyəh f √dhy very learned, very good (person). These cases are ambiguous: for fāydət, ML gives a variant with a short a, k’ēʕyōt presents a quite exceptional ʕ on the surface (Rubin 2010: 15), ħāgērtən at last is given by Jahn with a short vowel (ħeyjarten). Let us finally mention the case of the collective noun bɛ̄r (< √bʕr camel). In ML: xv, for the forms with possessive singular suffixes, Johnstone gives hə-bɛ̄r-hɛ, -sɛ, -kɛ etc. his, her, your (m) camels etc., i.e. forms with a long vowel in closed syllable. However ML s.u. and Rubin (forth.): e.g. 377, Text 10:11, give suffixed forms with a short vowel, as expected (e.g. hə-bɛr-kɛ your (m) camels).


Roots ʕbr, ʕðr, ʕθ’l, ʕθ’m, ʕgm, ʕgz, ʕkb, ʕk’d, ʕk’l, ʕk’r, ʕlg, ʕlk, ʕlm, ʕrb, ʕs’b, ʕwr.


We shall not consider the -vh# / -vʔ# variants of #, e.g. məlitoh for məlitō (ML: xxiv n. 1, xliv n. 2, lii n. 2 and ML pass.). This transcription encodes a prepausal devoicing/breath in the realisation of long vowels. This effect obtains at the phonetic level, not at the phonological level.


The vowel in the following syllable surfaces as an unstressed a instead of ə. This is a regular effect of the presence of /ʕ/.


We note a few triliteral forms—often borrowings—with final CVC#: √nfr nəfar (Ar.) individual, person, √šnb šənab (Ar.) short moustache, √θbt θəbat a kind of draughts (the syllabification of the M word indicates that this is a borrowed word), √š’bʕ š’əbaʔ (also š’abaʔ) finger, toe, √s’nb s’ənab idol, √glћ gəlaћ cuckold. These few forms are absent in our recordings, we leave the question open.


This result is in line with Watson (2012: 17): “acoustic evidence from fricatives and released final stops provides strong indication that final geminate verbs do indeed end in final geminates phonetically, and that these contrast in length with simplex counterparts” and with Rubin (forth.). In other terms, today Omani Mehri shares phonetically realised final geminates with Yemeni Mehri. It is difficult to know if Johnstone’s transcriptions with a simplex final C are inadequate or if the pronunciation of these forms has evolved between the time when he collected these data and our own data. In this last case, in the forms produced by Johnstone’s informants, the geminate would be present at the phonological level, but not at the phonetic level (Bendjaballah & Ségéral 2013).

From the time when Johnstone recorded his data, the situation would therefore have evolved towards an obligation of phonetically realising the geminate. We find a similar evolution in an isolated verb, tək’ drink < √hk’y. This verb presents an underlying final geminate which is realised on the surface by our informants: [əttək’k’] pf 3ms “he has drunk”. This evolution can generally be observed in the language, word-finally as well as word-initially, see Bendjaballah & Ségéral (2017) for more details.


In ML, the duals of “–ǝn forms” are systematically ending with -ayǝn with one exception: yǝhǝgārēyǝn h-F √gry put forward, advance (ML: lxii) which has a long vowel.

Furthermore, the feminine of adjectives in -ət has a morphological structure similar to that of the duals of “-ǝn forms”. In this case, Johnstone transcribes the suffixed form with yy: məhray, f məhrayyət Mehri √mhr, ħəbəɬay, f ħəbəɬayyət Ethiopian √ħbɬ etc. Perhaps the fact that the geminate is systematically missing in the duals reflects a systematic convention, and not the transcription of the attested forms?


The existence of two sequences, -vyyǝn for the reflexive, the reciprocal and the duals on the one hand and -yǝn for √CCv pf 1p on the other hand, is not in itself surprising. As a matter of fact, the morphological structures of these sequences are very different. In particular, in the pf 1p, y is a radical consonant.


The underlying vowel of arōkəb is /a/, like in yarakbən. For the sake of clarity, we keep surface o in (31). On *ā > ō, see Johnstone (1975: 11), Rubin (2010: 16).


According to Cantineau (1950: 107–108), a similar situation obtains in Biblical Hebrew: “Dans les syllabes ouvertes (parmi lesquelles il faut ranger les syllabes accentuées finales fermées par une seule consonne, car dans un état de langue antérieur ces syllabes étaient ouvertes par suite de l’ existence d’ une voyelle finale), […]”

Let us mention here that this final latent vowel could be the ī or ū that appears to the right of pf 3ms forms at the outcome of the concatenation of Obj morphemes: rəfūs pf 3ms √rfs kick + h Obj 3ms > rəfsīh, + k Obj 2ms > rəfsūk. This seems to be Johnstone’s interpretation: “It will be seen that the suffix is preceded by -ī- in the 3ms. perf. verb. It can be presumed that this, and the -ū- which other verbs have, derives from a final -a which no longer occurs in pausal forms” (ML: xvii). We will not go any further on this particular point: the cliticisation of Obj morphemes is a very complex question whose phonological implications must be more thoroughly studied (see Bendjaballah & Ségéral 2014b, Shlonsky this volume).


ə# only appears exceptionally: to our knowledge, only in N ðōblood and Fa ipf yəðōbleed.


The case of Obj 1s -i# is particular. When it is suffixed to pf 3fs, the preceding stressed vowel is short: wəzməti she gave me, *wəzmūti. This can be understood only if the morpheme is underlyingly /y#/ and not /əy#: wəzməti (< wəzmūt + y) is parallel to pf 3fs + Obj 3fs wəzməts (< wəzmūt + s) she gave her, to pf 3fs + Obj 2ms wəzmətk (< wəzmūt + k) she gave you (m), etc.


As indicated in 2.2.1, Johnstone transcribes u# as ⟨əw⟩ both in nouns—to the sole exception of təmbōku √tmbkw tobacco (MT 48: 13, 94: 26)—and in verbs: 1) Nouns: ɬētəw √ɬtw winter (January–March), ħēk’əw √ħk’w belt, ʁērəw √ʁrw one day old camel, dōləw √dlw well-bucket; well-rope, gōləw √glw fever, rēkəw √rkw red leather mattress of dyed cow-hide, 2) Verbs: Fb pf gēləw √glw be ill, fevered, pf kītəw √ktw (darkness) fall, pf s’aynəw √s’nw be deaf, h-F sbj yəhēχəw, part ms məhēχəw √χwy send a confidential message, sbj yəhīk’əw √k’wy build up (a fire); strengthen, sbj yəhēhəw √hwy make fall, sbj yəhēnəw √nwy intend, decide (to do st.), sbj yəhērəw √rwy give to drink, sbj yəhētəw √twy cause to eat, sbj yəhaɬ’ɬ’əw (a strange form seemingly deriving from a bilitary root √ɬ’w) √ɬ’wʔ light, š1-F impératif s šēləw make a turn-off, 4a-F pf fas’rəw, sbj 3fs təfas’rəw √fs’rw bring forth a caul (exhaustive list < ML). The recordings are unambiguous: they have [u]. A form like gēləw (MT 38.1; 7.9; 25.17; 84.2, 8; 96.3) always corresponds to [gīlu].

We finally must add two verbal forms for which ML gives a stressed ō# (transcribed as ū# by Stroomer in MT): Fb pf 3fp bǝrō (berū MT 85: 31) √brw bear, give birth (to) and 4a-F sbj 3ms yaʁsərō √ʁsrw chat at night (1p naġaserū MT 48: 29). These are mistakes of ML: the forms are in fact bīru for /bīrəw/, naʁasru for /naʁasrəw/ (Aaron Rubin p.c.).


The preposition (ə)nχāl- under functions like a N p regarding the suffixation of Poss (ML: xviii and s.u.).


Cf. section 3.1.


Johnstone (ML: xv n. 1) considering the ɛ# of the Poss suffixed to N p (35b), only examines the possibility of a vowel: “The ɛ of the final syllable of the sing. pers. suff. was doubtless long but unstressed”.